The AI Skills Gap Is a Measurement Error, Not a Deficit

On September 11, Disrupt Africa reported on a World Bank finding that global vacancies requiring generative AI skills increased ninefold between 2021 and 2024. The piece frames this as a problem for education systems everywhere, and a particular problem for Africa, where the article suggests the continent might need a different model because its schools cannot redesign fast enough to keep up. That framing has a hidden assumption worth pulling apart before anyone building here accepts it.

The assumption is that somewhere, a credential pipeline was working before generative AI arrived and is now failing to keep pace. Nine years of stable curriculum design turned out to matter less than eighteen months of tool release cycles. That is a real story, but it is a story about disruption to something that existed. In Kampala or Nairobi or Kigali, the pipeline being disrupted was never as entrenched in the first place. Fewer institutions built their hiring, their salary bands, and their promotion ladders around a fixed technical curriculum that assumed a ten-year shelf life. That is not a strength born of foresight. It is an accident of never having built the thing that is now breaking everywhere else. But an accident that removes institutional inertia is still an advantage, and it is one that rarely gets named in coverage written from outside the region.

What credential are you actually hiring for

Systems that answer questions and systems that do work are different categories, and most organisations buy the first while believing they bought the second. The distinction matters here because the skill that makes someone useful next to a system that answers questions is different from the skill that makes someone useful next to a system that books appointments, reconciles payments, or updates a booking calendar without a human checking every output. The first skill is comfort with a chat interface. The second is the ability to notice, at 4pm on a Tuesday, that a workflow has quietly started producing wrong answers with high confidence, and to trace why before it reaches a customer. No degree program anywhere issues a certificate for that. The World Bank's ninefold demand increase is a demand signal for the first skill. It says nothing about who can supply the second, and the second is the one an agentic studio actually needs.

This is where the infrastructure race, however real, is a separate question from the talent question. Nigeria's UduTech signed a GPU supply deal with South Korea's Baro AI on September 9, expanding compute access for African governments and businesses, according to TechCabal. Compute matters, and it will keep mattering. But a business with abundant GPU access and a hiring process calibrated to formal AI credentials will still deploy a system that fails quietly in the field, because the bottleneck was never the chip.

The objection that matters

The strongest case against this argument is that formal training still filters out people who cannot reason at all, and that skipping the credential leaves you with unfiltered volume rather than talent. There is a version of this concern that has real weight for education specifically. TechCabal covered a related worry about students in overcrowded African classrooms coming up through school with instant AI answers replacing the productive struggle that builds actual reasoning, and about what that means once those students reach the workforce. If a generation of graduates arrives having outsourced the struggle to a chatbot, the fear is that the diagnostic instinct an agentic deployment needs never gets built at all, credential or no credential.

That objection is worth taking seriously, and it argues for something more specific than it first appears. It is not an argument for trusting formal credentials over demonstrated judgment. It is an argument for testing judgment directly rather than inferring it from either a degree or the absence of one. A transcript tells you nothing about whether someone can sit with a broken workflow and figure out where it broke. Neither does the fact that someone grew up with a chatbot in their pocket. The only way to know is to give them a system that is actually failing and watch what they do. That test is more expensive to build than a resume screen, and it is the only one that has ever produced a reliable signal, in any labour market, for this specific kind of work.

The practical shift for anyone hiring into an agentic build here is to stop treating the absence of a formal AI credential as a gap to apologise for. The credential does not exist in a form that would help you anyway, not in Boston and not in Entebbe. What exists, unevenly distributed everywhere, is people who can diagnose a broken process. The region's disadvantage was never a shortage of degrees. Its advantage, rarely stated out loud, is that fewer of its hiring pipelines were ever built around the assumption that a degree would tell you.

Sources

AI is changing work faster than education can adapt, Africa may offer a different model (Disrupt Africa, 2026-09-11)

Nigeria’s UduTech taps South Korean GPU supplier to expand Africa’s AI compute (TechCabal, 2026-09-11)

Are we raising kids who can’t think without an AI prompt? (TechCabal, 2026-09-10)